Help
Hi again,
Am also frustrated when Gib doesn't lead the suit you have bid and raised. Instead on game #6600 hand #6 he leads trumps and pickles a possible trump trick. Thanks
Page 1 of 1
Gib leading suit you have bid and raised
#2
- Group: Advanced Members
- Posts: 6,542
- Joined: 2010-February-01
- Gender:Male
- Location:Brooklyn, NY, USA
Posted 2012-October-30, 09:54
#3
- Group: Advanced Members
- Posts: 4,155
- Joined: 2003-May-14
Posted 2012-October-30, 11:50
It's pointless to complain about stuff like this. You have to remember that GIB doesn't think at all like a human, ever. It is dealing a bunch of hands that fit the auction and picking the best card on average on a double-dummy basis. It is not going through human maxims like "lead partner's suit, lead from a sequence, etc." You may think that you want them to rewrite GIB to use human type rules to select a lead, but you would be wrong, it would almost certainly play much worse on average! Programs prior to GIB, early 90s, tried to mimic human thought process, play by "rules". GIB changed to this "Monte carlo sampling" technique and crushed these earlier programs, and everyone switched to the simulation/brute force technique. Plus it's not feasible for them to change the play/defense engine anyway, given their current development resources.
GIB actually probably does better on leads on average than humans, on a double-dummy basis, in my estimation. Computer sim probably does better than human leads which are often guesswork even for experts.
Also on this hand, the trump lead gave up nothing on a double-dummy basis. On some hands, this trump lead might be critical to prevent E-W from ruffing a club.
The only thing you can really get them to change regarding leads is *systemic* leads, which cards to lead from an honor sequence (in theory they could get GIB to say play Rusinow which would be silly since so non-std), or to play 3rd/low leads instead of 4th, or what card to lead from three small in various circumstances.
If they ever get the programming resources to improve GIB's defence, the main things they have to do are teach it to signal and read signals, and to simulate single-dummy declarers instead of double-dummy so it doesn't do stuff like assume declarer is dropping its honor offside or playing for hook/deep hooks instead of drop & randomly popping honors to make declarer's life easier. The opening lead itself is relatively OK.
GIB actually probably does better on leads on average than humans, on a double-dummy basis, in my estimation. Computer sim probably does better than human leads which are often guesswork even for experts.
Also on this hand, the trump lead gave up nothing on a double-dummy basis. On some hands, this trump lead might be critical to prevent E-W from ruffing a club.
The only thing you can really get them to change regarding leads is *systemic* leads, which cards to lead from an honor sequence (in theory they could get GIB to say play Rusinow which would be silly since so non-std), or to play 3rd/low leads instead of 4th, or what card to lead from three small in various circumstances.
If they ever get the programming resources to improve GIB's defence, the main things they have to do are teach it to signal and read signals, and to simulate single-dummy declarers instead of double-dummy so it doesn't do stuff like assume declarer is dropping its honor offside or playing for hook/deep hooks instead of drop & randomly popping honors to make declarer's life easier. The opening lead itself is relatively OK.
Page 1 of 1